Google will let you generate a birthday song in the Gemini app now. It won’t tell you whose music taught the model how.
That’s the shape of the Lyria 3.5 release: a music generation model shipping straight into the Gemini app and through the API, with a pointed claim about training data attached and no receipts behind it.
The licensing claim, and the hole in it
Google says the Lyria model was trained only on licensed content, and it draws the contrast directly with Suno’s music model. That’s a swipe at the most obvious competitor in generative music, and it’s the sort of line a company only makes if it thinks it’s on solid ground.
But Google hasn’t shared any specifics about what training data it actually used. No catalogs, no labels, no rights holders. You get the assurance and none of the evidence, which means the claim is worth exactly as much as your trust in the company making it.
What you can do with it in Gemini
Inside the Gemini app you pick a genre and a style. You choose between vocals and instrumentals. You make a short track or a longer one.
Google is also shipping templates, aimed squarely at people who’ve never prompted a music model before: background music, or a personalized birthday song. Those are starter rails, not the product’s ceiling.
What’s changed since the last model
Google’s pitch for 3.5 over its predecessor comes down to two things: more expressive vocals and richer musical arrangements. That’s the company’s own characterization, and it’s the kind of improvement that’s hard to argue with and equally hard to verify from a spec sheet. Expressiveness isn’t a number.
It’s not just the Gemini app
Lyria 3.5 also lands in Google Flow Music, where Google says it comes with added features. Developers get it in Google AI Studio. And it turns up in Google Vids, which puts generated music inside the workplace video tool rather than a creative sandbox.
That spread matters more than the model update itself. A music generator in a consumer chat app is a novelty. The same model wired into a video editor and an API is Google trying to make generated audio the default fill for anyone who needs 30 seconds of something under a slide.
If you’re deciding whether to use it, the practical question isn’t whether the vocals sound better than last time. It’s whether you’re comfortable putting audio into published work when the company behind it says the training data was licensed but won’t name a single source.
